David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience

Ineffable Intelligence aims to build a "super learner"—an AI system capable of autonomous discovery that surpasses human knowledge without relying on massive human-generated datasets. By prioritizing reinforcement learning over language-based models, the lab seeks to move beyond the limitations of current LLMs, which often incorporate human biases and require extensive human-in-the-loop intervention. Founders David Silver and Lasse Espeholt, known for their foundational work on AlphaGo and scalable reinforcement learning, utilize high-fidelity simulated environments called "crucibles" to foster emergent complexity. This approach treats superintelligence as a scientific challenge, emphasizing the need for agents to learn from experience rather than mimicking existing human data. The team maintains a research-led culture, leveraging massive compute clusters to accelerate discovery while committing to rigorous safety gating before any real-world deployment, aiming to achieve superintelligence within three years.